There is a warning in red letters in galaxy manual:
"You should NEVER EVER use RPKM, FPKM, or TPM to compare expression levels across samples. These are RELATIVE measures! Consider yourself warned!"
Why? If I want to compare gene X between samples, and TPM normalizes to total reads, shouldn't it be ok to say gene X increase between samples 1 and 2? If someone can direct me to a clear explanation that would be very appreciated